previtus/OxHyperMinerals_Train

OxHyperMinerals dataset

Author: Vít Růžička

One of the OxHyper datasets, for more details please check my main web at: https://previtus.github.io/

Fast data preview in: https://huggingface.co/datasets/previtus/OxHyperMinerals_Train/blob/main/dataset_exploration.ipynb

The "OxHyperMinerals" is a completely new dataset for mineral identification, with labels of 3 selected mineral classes and source annotation for all 381 constituents (separated into two groups). In total we have 796 large tiles in the training dataset, 198 in the validation and finally 200 in the test dataset. These large tiles exactly match the tiles used in the synthetic methane events dataset (https://huggingface.co/datasets/previtus/OxHyperSyntheticCH4), however they do not contain the simulated events and contain more bands. The total size of this data is 372 GB. It contains all 285 spectral bands in between spectral range of 381-2492nm. We have aggregated the constituents released in the EMIT L2B data into selected mineral groups and kept them as binary classes. Please note that these labels may be experimental only - as such, we also provide the source EMIT L2B labels for each location.

More info will be added upon publication, for now please refer to https://previtus.github.io/

More information about the NASA's sensor EMIT: https://earth.jpl.nasa.gov/emit/

Splits

Train (~248 GB): https://huggingface.co/datasets/previtus/OxHyperMinerals_Train

Val + Test (~124 GB): https://huggingface.co/datasets/previtus/OxHyperMinerals_TestVal

It is suggested to download both repositories and to place the data into the same folder (the csv files will allow the Train/Val/Test split).

Note: The Train repository will contain the up-to-date information (TestVal will likely be kept as it is)

File structure

Each folder represent a large tile (of 512x512px). Rows in the attached csv's split these into train/val/test subsets.

Each folder contains these files:

• C - all 285 bands in a georeferenced file (note: you'll also need the C.hdr file to load it, can be also inspected for example in QGIS)

• minerals3ghk.tif - 3-class label with minerals of Goethite, Hematite and Kaolinite

• mineral_ids, band_depths - source data for all 381 constituents separated into two groups, please refer to the Theoretical Basis document for EMIT L2B (https://lpdaac.usgs.gov/documents/1659/EMITL2B_ATBD_v1.pdf) for more details

• B_magic30_tile.tif - computed matched filter product (using the iterative variant called mag1c)

• info.json - saved information about the used source data (for example the link to the used EMIT_L2B_CH4PLM sample)

• Note 1: for each georeferenced tif we also provide the .hdr file which is needed for loading and has details about the bands

• Note 2: The RGB bands are not provided in this case (as we didn't need them extracted for a RGB only model)

• Note 3: Files minerals3kmc.tif (with labels for Kaolinite, Illite+Muscovite, Calcite) and minerals10.tif (with the full set of [Calcite, Chlorite, Dolomite, Goethite, Gypsum, Hematite, Illite, Kaolinite, Montmorillonite, Vermiculite]) are added to enable future experiments.

Citation

Will be added.